Observed on v0.1.0 (latest release) / main @ 0b6fad6; dogfooded 0.1.1.dev against linopy's suite.
linopy has two sweep kinds in one suite: models over size n (10–10⁴) and patterns over a 0–100 severity. Both land in the numeric dim, plus a categorical axis ("n" / "severity") that distinguishes them. plot_scaling over a mixed snapshot then puts sizes and severities on one log x-axis, squishing the patterns at the left — the auto-infer can't know the two numerics are incommensurable.
Proposed: when a categorical dim partitions the numeric into disjoint ranges (or simply when an axis-like dim exists), auto-facet by it instead of overlaying — or at least warn. More generally: a heuristic so plot_scaling doesn't place values from different units on a shared axis.
Acceptance: a snapshot mixing two sweep axes scales each in its own facet by default.
🤖 Issue drafted by Claude from the linopy integration.
Observed on v0.1.0 (latest release) / main @
0b6fad6; dogfooded0.1.1.devagainst linopy's suite.linopy has two sweep kinds in one suite: models over size
n(10–10⁴) and patterns over a 0–100severity. Both land in the numeric dim, plus a categoricalaxis("n" / "severity") that distinguishes them.plot_scalingover a mixed snapshot then puts sizes and severities on one log x-axis, squishing the patterns at the left — the auto-infer can't know the two numerics are incommensurable.Proposed: when a categorical dim partitions the numeric into disjoint ranges (or simply when an
axis-like dim exists), auto-facet by it instead of overlaying — or at least warn. More generally: a heuristic soplot_scalingdoesn't place values from different units on a shared axis.Acceptance: a snapshot mixing two sweep axes scales each in its own facet by default.
🤖 Issue drafted by Claude from the linopy integration.